DEV Community: Sakeena Patel

DeadLock: The Ghost in the Local Machine 🎃👻

Sakeena Patel — Sun, 01 Nov 2026 05:24:02 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine.

What I Built

Imagine you’re locked in a server room after hours. The lights are dead. The terminals are humming. But the monster haunting the dark isn't a zombie or a vampire it's a silent, insidious AI engineering bug.

DeadLock is a local, browser based puzzle game where your weapons are debugging skills and your enemies are real-world AI failures.

Across 5 eerie levels and 10 distinct challenges, you inspect a AI engineering challenges edit it right in your browser, and hit Compile and attack.
Treat 🍫: Your code passes the assertions. You survive.
Trick: Your code fails. No points, no escape, try again.
A flawless run nets you 50 points. Need help? You can summon the ghost in the machine for a hint—powered entirely by a local open-weight model—at the steep cost of 5 points per clue. Two hints max for the whole game. Choose wisely.

Demo

DeadLock_Demo.mp4 - Google Drive

drive.google.com

Code

dataWizard7957 / deadlock

DeadLock

DeadLock is a local, browser-based game for practicing AI-engineering tasks. Players repair short Python functions covering prompts, sampling settings, context limits, retrieval, and evaluation. The puzzles run entirely on the machine. Hints, when used, are generated by a local model through Ollama.

How to play

The game has five levels, and each level has two challenges. A challenge is a broken function. Edit the code and choose Compile and attack.

A correct fix is a Treat. Clearing both challenges in a level awards 10 points.
An incorrect fix is a Trick. The score does not change, and the challenge stays open.
A perfect run with no hints, is 50 points.

The player has two hints for the whole game. Each hint costs 5 points and requests a short clue from the local model. The game remains playable if Ollama is not running; the hint then reports…

View on GitHub

How I Built It

The core gameplay loop is completely offline-first. To ensure the game stays snappy and playable even if your local LLM server is offline, the puzzle verification doesn't use an AI model at all. Every player fix is instantly validated via deterministic Python assertions running locally on the machine.
The open-source AI piece is reserved for the atmosphere. When you burn a hint, dungeon/dm.py sends a minimalist prompt directly to a local Ollama
It explicitly ignores proxy settings to ensure local routing stays local.
It optimizes for gemma2:2b if installed, handling cold starts gracefully.
It asks the model strictly for a clue, never the raw solution.
The frontend uses Streamlit to create a seamless, retro-arcade coding experience right in your browser.

Why Does Open Innovation Matter?

When you play DeadLock, you are writing snippets of real AI systems: prompt architecture, sampling parameters, retrieval strategies.
If this were built on a commercial chat API, every keystroke and hint request would siphon your code off to someone else's servers, running up a token bill while leaking code architecture.
A local, open-weight model flips the script:
Absolute Privacy: Your code never leaves your hardware.
Zero Marginal Cost: Hints cost game points, not API dollars.
Unfiltered Tone: The Halloween flavor and system prompt live locally, untouched by corporate safety filters or external policy classifiers.
Resilience: If Ollama drops connection, the puzzles still work. The ghost is optional. The engineering is absolute.

The Spirit of It

The magic of DeadLock lives in the space right beneath the code editor. When you click attack, the machine responds instantly with either a Treat 🍫 or a Trick.
There are no long-winded paragraphs explaining the bug above the code. You read the syntax, you spot the flaw, you fix the logic, and the machine either opens the door or locks you out. Pure logic meets midnight atmosphere.

Prize Categories

Best Use of Gemma

Used Gemma as hint powered entirely by a local open weight model.

2.Best Use of ElevenLabs

It was used for recording voiceover for demo.

Slow Tea, Steady Pay:

Sakeena Patel — Mon, 26 Oct 2026 06:02:47 +0000

Chill Pay ☕

What I Built

Ever since Dad started his startup, his mornings became business.

But the 1st? The 1st was something else.

One morning, the 1st, I woke up early. But guess who was already awake?

Dad. Laptop open. Calculator in hand. A cup of hot tea beside him, slowly going cold.

Is there an exam today? Why is everyone so tense?

Oh. Right. It’s the 1st. Salary day.

I can’t take his worries away. But I could make sure that on the 1st, at least, his tea stays hot. So I built Chill Pay, a local payroll and invoice app that turns his messy monthly notes into approved pay and invoice PDFs.

Dad can enjoy his hot tea. And any small startup owner who currently does payroll with WhatsApp messages, a calculator, and a lot of worry.

The problem

My father runs a small startup in India. Some people are monthly employees, some are contractors, and some are trainees. Every month, leaves, extra work days, and bonuses reach him as messages and scribbled notes. He turns them into each person’s final pay by hand. One wrong number means an awkward conversation with someone who is counting on that money.

Cloud payroll tools charge per employee and want his team’s salary data on someone else’s servers. For a small startup, that’s a hard trade.

What Chill Pay does

This month: Dad types or pastes his note. A local model reads the names and numbers out of it. He then reviews leaves, extra days, and comp-off, and code calculates every row amount.
People: members, departments, and projects. Each member is an employee, contract, or trainee, with their own pay.
Records: past months, invoices, reports (payroll register, monthly TDS, financial-year TDS, CSV export), and advances.
Setup: company details printed on invoices, login accounts, and Department Head leave.
Approvals: a Department Head approves or rejects a submitted entry with a comment. If a department has no head, or the head is on leave, HR can approve instead.
Invoices: each approved member gets one PDF, numbered from their name and billing month (e.g. RAKU_INV202609 for Raj Kumar, September 2026).
Roles: Finance sees Records only. Members and Department Heads see This month and Records. HR sees everything.

The rule I never broke: the AI never calculates money. It only reads human messages.

Demo

chill_pay.mp4 - Google Drive

drive.google.com

Code

dataWizard7957 / chillpay

Chill Pay

Chill Pay is a local payroll and invoice app. Each account is its own company. Notes for a month are reviewed by a person, then Python calculates the pay and creates one invoice per member.

The chat model only reads a note. It does not calculate money. Pay is calculated in app/services/payroll_engine.py.

What you can do

The side panel has four pages.

This month

Notes. Type a payroll note in chat, or paste it on the monthly run. A saved member can be matched by full name, or by a first name when only one member has that first name.
Review. Check leaves, extra days, and comp-off. The amount on the row is calculated by code.
Decisions. A Department Head approves or rejects a submitted entry with a comment. HR can approve a monthly run when that department has no head yet. After a head is assigned…

View on GitHub

How I Built It

Model: gemma:2b, an open-weight Gemma model running locally through Ollama
Frontend: React
API: FastAPI
Database: SQLite, one file on Dad’s machine

How the AI fits in:

the note goes to Gemma, which pulls out names and numbers. Those land in an editable review table, not straight into payroll. If Ollama isn’t running, Dad can type the note by hand and nothing breaks.

And python does math.

Guardrails for a “never wrong” app:

Months before a member’s joining date can’t be approved.
Approved payroll is saved as a snapshot, so later edits never rewrite history.
Each company login is its own world. One company can’t see another’s members, pay, or invoices.

The hardest part:

Getting a 2B model to return consistent structured output, handling phrases like "Ravi half day Tuesday", or deciding where to stop trusting the model and start asking the human.

Why Does Open Innovation Matter?

Privacy. Salary is some of the most sensitive data a company has. With a local open-weight model, nothing goes to a cloud model. The data, the calculations, and the invoices stay in a SQLite file on Dad’s own machine. He doesn’t have to trust anyone’s privacy policy.

Cost. No per-token bills and no per-employee fees. A tool like this only works for a tiny startup if running it costs almost nothing.

No lock-in. No vendor can change the price, the terms, or the API underneath him. If a better open model arrives next year, I swap it in.

A small model is enough. The job is narrow: read short, messy notes and pull out names and numbers. A 2B model on a normal laptop does that, and since code does the money, the model never has to be perfect, only helpful.

The Lazy Payoff

Slow and steady isn’t about doing things slowly. It’s about not rushing the parts that can’t be wrong. Chill Pay is deliberately unhurried where it matters:

A small local model that reads, but never decides.
Every row calculated by code, the same way every month.
A human review before anything is approved.
A snapshot, so the past stays put.

And that leaves the morning slow in the nice way.

Payroll used to take Dad about 4 hours. With Chill Pay it took 10 minutes.

His reaction: Wow, now it can done in minutes.

Prize Categories

Best Use of Gemma: gemma:2b runs locally through Ollama and turns messy payroll notes into names and numbers, with a manual fallback when Ollama is off.

Best use of eleven labs : It was used for voice over of demo video.

Running a startup and dreading the 1st of the month? ☕ Tell me which part of your workflow is the slow and messy. I’m taking on a few small pilot projects.

💬 Comment below or DM me and we can talk.

🌿Alt-nature:

Sakeena Patel — Mon, 12 Oct 2026 07:21:37 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

🌿Alt-nature: An AI-Powered Micro-Quest Platform for Screen-Free Breaks

We spend our lives staring at glowing rectangles, waiting for digital fatigue to burn us out. We know we need to step away, but walks feel aimless. Alt-nature solves this by turning the real world into an interactive, AI-driven adventure without forcing you to stare at your phone while you do it.

When you type in your location and pick an outdoor vibe like a Fall Foliage Walk or Forest Trail, the app fetches real-time local weather via SerpAPI, generates a personalized sensory quest using an open-weight Gemma model, crafts immersive environmental lore, and synthesizes a voice-guided mindfulness briefing using ElevenLabs TTS. Once you step outside, complete your walk, and return, you can log your reflection directly back into the platform, which stores your completed milestones in MongoDB Atlas.

Demo

GitHub Repository:
dataWizard7957 / Alt-nature
Alt-nature 🌿

Alt-nature is an open-source AI agent system designed to drag software engineers away from their screens and into nature. By combining open-weight model inference with environmental grounding , speech synthesis , and durable persistence , Alt-nature delivers screen-free outdoor audio quests.

Real outdoor exploration happens where cell reception dies. By leveraging open-weight models like Gemma, Alt-nature guarantees zero latency, absolute privacy for real-time outdoor locations, and complete independence from closed proprietary APIs. You get full control over your prompt pipelines and system guardrails without vendor lock-in.

🛠️ Tech Stack
- Core AI Model: Gemma (open-weight model for dynamic quest generation)
- Voice Narration: ElevenLabs API (for hands-free, screen-free audio briefings)
- Data & Persistence: MongoDB Atlas (for storing active quests, reflections, and user history)
- Environmental Grounding: SerpAPI (for real-time weather and seasonal status)
- Agent Observability: Sentry Agent Tracing (tracking execution telemetry, routing, and inference fallbacks)
⚙️ Quickstart
1. Clone and Setup…
View on GitHub
Screen recording:

Code

The project is split into a robust FastAPI backend and an interactive Streamlit frontend. Here is a look at the repository layout:

Alt-nature/
├── app/
│   ├── core/         # Database and configuration
│   ├── models/       # Pydantic validation schemas
│   ├── routers/      # API endpoints (quests generation & verification)
│   └── services/     # Gemma, SerpAPI, and ElevenLabs integrations
├── frontend/         # Streamlit UI application
└── static/           # Audio narration storage

How I Built It

Alt-nature relies on a modern, modular AI engineering tech stack:

Backend: FastAPI with Pydantic schema validation and asynchronous routing.
Database: MongoDB Atlas for securely tracking active and completed user quests and reflections.
AI & Generation: Gemma (open-weight model) for dynamic quest generation with local fallback support.
Environmental Context: SerpAPI to pull live local weather data and tailor quest objectives to current conditions.
Voice & Audio: ElevenLabs TTS API to synthesize immersive, calming audio briefings for screen-free guidance.
Frontend: Streamlit for rapid UI prototyping with custom emoji-coded themes and responsive audio players.

Why Does Open Innovation Matter?

Open innovation specifically leveraging open-weight models like Gemma was foundational to building Alt-nature. Closed, proprietary APIs often impose rigid constraints and strict rate limits that make real-time sensory prompt engineering and localized environmental blending difficult to customize. Open-weight models gave me the flexibility to fine-tune system prompts for mindfulness, run local test inferences during development, and build deterministic system guardrails without vendor lock-in.

Prize Categories

Open-Source AI Challenge (Week 1: Touch Grass)
Best Use of ElevenLabs: Alt-nature leverages ElevenLabs TTS to convert generated environmental lore into calming, screen-free voice briefings for outdoor walks.
Best Use of MongoDB Atlas: MongoDB Atlas securely manages persistent state tracking for active quests, user reflections, and completion statuses.
SerpAPI: SerpAPI fetches live local weather data to dynamically tailor micro-quest objectives to real-world environmental conditions.
Sentry: Sentry provides real-time error monitoring and agent tracing across backend FastAPI routing and inference fallbacks.

DEV Community: Sakeena Patel

DeadLock: The Ghost in the Local Machine 🎃👻

What I Built

Demo

DeadLock_Demo.mp4 - Google Drive

Code

dataWizard7957 / deadlock

DeadLock

How to play

How I Built It

Why Does Open Innovation Matter?

The Spirit of It

Prize Categories

Slow Tea, Steady Pay:

Chill Pay ☕

What I Built

The problem

What Chill Pay does

Demo

chill_pay.mp4 - Google Drive

Code

dataWizard7957 / chillpay

Chill Pay

What you can do

How I Built It

How the AI fits in:

The hardest part:

Prize Categories

🌿Alt-nature:

What I Built

Demo

dataWizard7957 / Alt-nature

Alt-nature 🌿

🛠️ Tech Stack

⚙️ Quickstart

Code

How I Built It

Why Does Open Innovation Matter?

Prize Categories